Metabolomics for the masses: The future of metabolomics in a personalized world

1Central to this review is the role of metabolomics within the clinical sciences and so metabolomics as a discipline is first introduced, along with the role of clinically useful biomarkers (small molecules). Following this we discuss metabolomics approaches for personalised and precision medicine and the future role of delivering metabolomics to the masses.

2Whilst there are many definitions of metabolomics we consider that metabolomics is a multidisciplinary science that seeks to define the entire complement of small molecular weight molecules termed metabolites within a biological matrix of interest. Metabolomics has been readily applied to a vast array of biological matrices of pre-clinical and clinical medicine relevance, with perhaps not surprisingly the most common being blood plasma and serum as well as urine. These are not the only samples accessible to the clinician and many studies have also focussed on extending these measurements towards intact tissues. This is particularly important for cancer diagnostics as measuring the pathology directly is likely to yield pathophysiological information about the disease (i.e. the cause) rather than measuring circulating metabolites (i.e. the likely downstream effect). In addition, studies have also shown that it is possible to generate information-rich metabolomes from human saliva, breath, cerebrospinal fluid (CSF), broncho alveolar lavage (BAL), sweat, faeces (as well as other locations in the gastro-intestinal tract), semen, and amniotic fluid. Finally, some research has also cultured primary cells for mammalian cell-based models, which may be particularly important for ADME-Tox (adsorption, distribution, metabolism and excretion-toxicology) studies.

3The term metabolomics was first coined in the late 1990s and had its 18th anniversary last year. Metabolomics has increased in popularity and applicability ever since. Metabolomics can no longer be described as a novel concept within the clinical arena and it is now emergent. A simple search of Web of Science (on 7th Feb 2017) for (metabolom* OR metabonom*) AND (clinical OR medicine) within the research topic field returns over 3700 articles. Within the range of ‘omic approaches (i.e. transcriptome, proteome) the metabolome is perhaps the most closely linked to the phenotype of the subject and thus, can report on disease status as well as the effect and response to external stimuli (e.g. drug therapy, nutrition, exercise, etc).

4To treat disease GPs, clinicians and health workers require diagnostic indicators of disease, which can be used not only to diagnose said disease but also to assess the applicability of therapeutic interventions. These indicators are referred to as biomarkers and the NIH definition of a biomarker is:“A characteristic that is objectively measured and evaluated as an indicator of normal biological processes, pathogenic processes, or pharmacologic responses to a therapeutic intervention.”

5“A characteristic that is objectively measured and evaluated as an indicator of normal biological processes, pathogenic processes, or pharmacologic responses to a therapeutic intervention.”

6The use of biomarkers to direct therapeutic intervention in terms of whether dietary and lifestyle interventions are necessary, or whether drugs are appropriate or if surgery is needed, is highlighted in Fig. 1.

7A summary of clinically useful biomarkers used in clinical practice are provided in Table 1 (summarised from a test catalogue provided by Mayo Clinic, US available at http://www.mayomedicallaboratories.com). Whilst these biomarkers are very valuable for diagnosing disease there is an urgent need to have many more that are highly predictive and robust. There are many more biomarkers waiting to be discovered and this is a very active area of research in metabolomics.

8With such popularity and regard, metabolomics as a discipline must look towards the future and begin to anticipate how the field can develop and transition into the modern-day world. As mentioned above metabolomics is now established, and whilst work is on-going in terms of technology and computational improvements, the method is now considered routine. Despite this maturity, the literature is unfortunately saturated with small-scale preliminary-type studies with many suffering from being poor in experimental design and thus any findings are likely to be false as they lack statistical robustness and validity. This is not a unique feature of metabolomics and as nicely exemplified by George Poste in his article entitled “Bring on the biomarkers”, whilst many biomarkers have been described in the academic literature comparatively few (well almost zero) have made it into the clinic. With reference to ‘omics and biomarker discovery one is often reminded of Henry Nix's famous statement:“Data does not equal information; information does not equal knowledge; and, most importantly of all, knowledge does not equal wisdom. We have oceans of data, rivers of information, small puddles of knowledge, and the odd drop of wisdom.”

9“Data does not equal information; information does not equal knowledge; and, most importantly of all, knowledge does not equal wisdom. We have oceans of data, rivers of information, small puddles of knowledge, and the odd drop of wisdom.”

10With this in mind we believe that the field must drive forward towards the undertaking of large cohort multi-centre studies to enhance the discovery of biomarkers that have increased prospects of translation into point of care and rapid diagnostics; this biomarker discovery process is highlighted in Fig. 2, and of course is not limited to metabolites but any molecule.

11Table 2 highlights several key metabolomics studies that have been aimed towards identifying biomarker candidates for an array of diseases. This table indicates the target disease of interest and the publication year, which illustrates the attempts made for biomarker discovery using metabolomics approaches, for a specific condition. It also summarises the number of control candidates and the number of diseased patients that were incorporated into the studies. Although these and other authors do not deliberately eschew obfuscation these numbers are often difficult to distinguish clearly within a manuscript. In addition, in some cases longitudinal studies are conducted whereby a patient is their own control. In order to have clarity in what was done within a study and what should be reported the Metabolomics Standards Initiative (MSI) initiated and subsequently published a series of papers on minimum reporting standards. Within Table 2 the biomarker (or biomarker panels) that have been discovered within each study are documented and, we note if an independent validation has occurred within the same study which will of course increase confidence in the validity of said biomarker.

12It is clear from inspecting this Table 2 that there is a broad difference in the number of subjects included in these studies. The community is yet to decide what this number should be, but it should be noted and acknowledged that the availability of patients will greatly vary from disease to disease and equally access to valuable (sometimes very rare) samples will be limited. In this century alone, there have been more than 1600 publications (using a combined search of the above PLUS biomarker* from 2000 to date) that ‘claim’ to have discovered a biomarker using a metabolomics approach, which is nearly half of all papers surveyed! Although there are some exceptions, most of this research fails to acquire enough statistical power due to a limited sample size (<100 subjects in total) and almost none repeat the analysis in a further cohort and thus fail to demonstrate a lack of biomarker utility. We believe that these thwart the potential translation of metabolomics research into clinics. For instance, there is minimal-known translation of metabolomics biomarker discovery into clinics for the top five causes of death in the UK (Table 3) which include: ischaemic heart diseases, dementia and Alzheimer's disease, malignant neoplasms of trachea, bronchus and lung, chronic lower respiratory diseases and cerebrovascular diseases. Malignant neoplasms, respiratory disease and ischaemic heart diseases are also three of the top five leading causes of death across Europe.

13Despite the above disease being of obvious importance we note the rapid rise of microorganisms as contributing to world-wide mortality. The obvious ‘culprits’ here being Mycobacterium tuberculosis and HIV, but with the almost meteoric rise in antimicrobial resistance (AMR) many normally harmless opportunistic pathogens will become increasingly important. Indeed it is predicted by 2050 that bacterial infections will kill more humans than cancer and heart disease. Whilst it is accepted that there are many microbial interactions with the host cell microbiome and that man is a true superorganism it is also notable that many common human disease may indeed have a microbial origin. Metabolomics is likely to play a valuable role in understanding AMR and the host-pathogen interaction.

14This review seeks to provide an overview of metabolomics in respect to diagnostic applications and demographic screening and present a futuristic perspective on the implementation of the field with novel portable and wearable technologies.

15Although personalised medicine is a generic entity relatively new to the field of healthcare research, it has of course been practiced for decades within a so-called evidence-based framework (Fig. 3). In evidence-based medicine an individual is treated for disease largely based on the most popular medicine. After the drug is taken for some time an assessment is made, with the desire to evaluate whether this has relieved symptoms (this may involve the measurement of a clinically useful biomarker (Table 1)). Based on this deterministic assessment the patient may then stay on the same drug, be diagnosed an alternate medicine, or be given a treatment to relieve side effects of the first drug. This process is slow and potentially dangerous to the patient. A much more desirable approach is to use precision medicine and this was brought to the forefront of attention when, during his 2015 State of the Union address President Obama announced that he was launching the Precision Medicine Initiative. This was heralded as a bold new research direction with changing for biobank made available by NIH to support the initiative.

16Precision medicine involves assessing the genotype (e.g. SNPs) and phenotype (e.g. metabolome) of the patient before they undergo any treatment (Fig. 3) and therefore relies on accurate analytical methods for directing therapy. Biomarkers are needed that can accurately identify the underlying pathology as these may help understand the disease aetiology and thereby result in a precise treatment,. Clearly the lack of suitable biomarkers currently holds back the wider implementation of personalised medicine. This is where metabolomics plays a key role as an approach to discover a biomarker, trial its detection within a large diverse population and then translate its detection into cheaper, quicker and reliable methods that could be used by a wider audience. As indicated above the main use of metabolomics as a tool is for biomarker discovery,,,,. The closest representation of a disease phenotype is a key-driving factor for the increased use of metabolomics for biomarker discovery to understand disease pathologies and finding methods of cure, and as many diseases result in changes in human metabolism it makes sense to use a method that measures metabolism directly!

17However, the focus of biomarker discovery should not only be for pathological cures but also for preventive screening of healthy individuals (Fig. 1), as earlier biomarkers may be useful in directing dietary and lifestyle changes prior to more radical surgical treatment. Within biomarker discovery this raises the tantalising idea that all healthy individuals should undergo some biomarker screen well before any disease is found so that any change in a biomarker(s) level is personalised; for example, someone with an already raised PSA level may not have prostate cancer and this higher PSA levels maybe indicative of an enlarged prostate as one ages. Offering a well-designed screening program at a reasonable cost may not always be possible due to the numerous associated challenges; these include monetary limitations (labour and consumable costs) as well as ethical, legal and social considerations for an opt-in test. The risk-benefit ratio needs to be clearly defined per disease for a successful personalised screening. Encouraging biomarker discoveries from within the plasma metabolome, serum metabolome,,,, urinary metabolome,,, as well as the volatilome,, show immense potential of dramatically reducing health risks in fatal health conditions such as cancer, congenital disease, heart diseases and respiratory diseases (Table 3). It should be noted that a key hurdle for translation of these biomarkers into a routine clinical test is the failure to validate. In the absence of a universally accepted procedure for metabolic profiling used for biomarker discovery, different sites use their own optimized procedures. Additionally, even if identical analytical platforms and routines are used, the inherent inter-laboratory variation will play a great role in detracting from the validity of a potential biomarker and there can never be certainty that the entire metabolome has been profiled with that said platform, and in fact it is accepted in the metabolomics community that there is no magic tricorder that measures everything. Thus, there is always a potentially ‘better’ biomarker waiting to be discovered.

18Some notable large scale and/or multicentre metabolomics studies have been successfully conducted (Table 2) to map the human serum metabolome, to identify biomarkers for incident coronary heart disease, and to study the response of Aspergillus nidulans to epigenetic perturbation with a hope to expedite the search for new pharmaceutical leads. A correct balance needs to be considered between large scale vs. small sub-population focused studies where the risk is minimal but with maximum benefits,. Due to the higher cost and effort involved in the analysis of samples by a standard metabolomics workflow, it is often tempting (albeit one could say lazy) to use a smaller sample size for biomarker discovery and pre-validation. However, such studies which lack the required statistical power for confident biomarker assessment will entice anyone to start designing specific assays for assessments in large cohorts (Fig. 2).

19Like all ‘omics which are data rich, metabolomics on humans is influenced by many confounding factors such as age, gender, ethnicity, diet etc. and thus, large validation studies with suitable control cohorts must be used to remove any potential bias,. Certain metabolites that alter with normal physiological changes may also be significantly different in a metabolomics study. By way of an example, citrate has been shown to increase with age even in healthy individuals. A recent metabolomics study indicated amongst other metabolites that citrate was a significantly important biomarker for cancer. However, since an increase in citrate could also be attributed to difference in mean age (17 cancer patients = 70 and 21 healthy controls = 60) rather than altered TCA cycle in cancer, in the absence of closely age matched case-control cohort such results need to be taken with caution before inferring pathological importance of such a biomarker.

20Whilst the current perception is that screening large control groups of healthy individuals at the same time as diseased populations is not an option for validation studies, this position must change. Indeed, many people already use wearable technology for the assessment of their exercise levels, heart rate, blood oxygen levels, as well as sleeping patterns, so collecting data on ‘healthy’ individuals is not that maverick.

21With recent technological advancements in the form of affordable hardware (e.g. pedometers which include heart rate monitoring), health apps on smartphones, fitness bands and smart-watches, it is feasible to generate large amounts of useful health-related data even in healthy populations,. These measurements are readily available on a personalised level and could be used to complement clinical studies. For example, in treatment regimens which may include nutritional and exercise advice.

22The tantalising question is whether metabolomics could be delivered to the masses on a personalised level? Whilst mass spectrometry linked to chromatography is a very power metabolomics platform for biomarker discovery, it is laborious and expensive and therefore unlikely to be suitable for large-scale screening of very large populations (i.e. when n > 10,000, which is of course still small when we consider that the earth's population is estimated to be >7.5×109; http://www.worldometers.info/world-population/). Of course, once a series of biomarkers are discovered and validated the scenario is different where one now knows the measurands and these can be detected and quantified using analytical chemistry. These can include methods based on:Lateral flow devices – much like the pregnancy test which is based on antibody detection of the appropriate antigen (viz., human chorionic gonadotropin (hCG));Dipstick approaches – for example the detection of nitrite for confirming urinary tract infections;Breath measurements for volatiles – for example ethanol detection and quantification using fuel cells for road side testing;Electrochemical detection – under skin glucose test is based on this and allows constant assessment of blood glucose that can be linked automatically to insulin injections.

23Lateral flow devices – much like the pregnancy test which is based on antibody detection of the appropriate antigen (viz., human chorionic gonadotropin (hCG));

24Dipstick approaches – for example the detection of nitrite for confirming urinary tract infections;

25Breath measurements for volatiles – for example ethanol detection and quantification using fuel cells for road side testing;

26Electrochemical detection – under skin glucose test is based on this and allows constant assessment of blood glucose that can be linked automatically to insulin injections.

27With the above in mind emerging technologies in metabolomics provide new platforms for high-throughput, highly sensitive, functional assays, biomarker discovery and offer opportunities for personalised medicine, complementing existing and emerging genomic, proteomic and transcriptomic technologies (Fig. 4). However, personalised medicine in the future could be better served when these biomarkers provide enough knowledge to translate them successfully into one or more types of wearable technologies that are readily available to an end user (as also illustrated in Fig. 4). Biosensors used in wearable technologies like smartphones,, smart-watches for monitoring heart conditions, health bands, necklaces, glucose monitoring contact lenses,, headbands etc., are excellent innovations transferring biomarker discovery onto a more individual level. Technological advances translating biochemical changes into physical signals is not something new, but in this age of bionics and biohacking, it is putting the technology in the hands of end user and thus able to boost the personalised medicine movement. Although, we recognise that these devices should always be with continued consultation with a clinical practitioner who can advise the wearer; much like home testing for blood pressure is currently practiced.

28Metabolomics studies seldom lack information about the studied bio-system, and although this science is described as data rich, as many metabolites are measured, it often lacks a sizable population. Being able to translate a subset of metabolite measurements onto a set of devices that not only validates the results but also provides other useful complementary information about the patient we believe is the next step forward.

29Chemometric-based analyses following any metabolomics study can immensely benefit from such wealth of continual metabolite data and metadata obtained from a target cohort. When metabolomics for the masses does occur the data processing may need to depend on large computing power and data storage space and these could be stored securely and privately within an advanced cloud-computing environment. In such a scenario, these measurements from different populations could be linked within the internet (so called internet of things (IoT)) allowing ensemble computation and for example epidemiological assessment of disease progression and spread (Fig. 4). This is not novel as the added benefit of current wearable biosensors is that the collected data are directly synced onto intelligent cloud services like Microsoft Azure or Google Cloud. An individuals (health) data collected over time via portable (a set of wearables) devices has the potential of producing copious amounts of telemetry data that can be computed over the cloud, producing predictions for future health risks. For example the mPower, mobile Parkinson's Disease (PD) study that attempts to research the occurrence, presentation and management of PD symptoms via survey telemetry data using a smartphone app. Another nice example is a smart-phone based application to monitor the association between pain and the weather for people suffering from rheumatoid arthritis. Use of artificial intelligence (AI) for screening, decision making and management whilest not new, can be adapted for risk stratification, prevention and choice of treatment in the healthcare systems. Recently, US Food and Drug Administration (FDA) has approved an AI based machine learning application for the use in clinics for making informed decisions about the health of heart where AI provides accurate measurements of the volume of each ventricle to physicians. This is thought to speed up and improve the decision making for heart surgeries. There is an enormous potential for a happy marriage between metabolomics and AI or machine learning technologies that are driven by data. This is where metabolomics should aim to take personalised medicine to - not only being able to predict a persons current or near future health or globally screen for potential biomarkers - but to link that information to dynamic metadata from patients to predict further risks and disease prognosis (Fig. 1). This approach as opposed to evidence-based medicine (Fig. 3) will enable better health care outcomes instead of trial and error treatment regimes.

30A potential future scenario illustrating precision medicine were together the patient and physician are at the centre of the diagnostics is shown in Fig. 5, once the hurdles of costs, barriers to patient inclusion and ease of use are overcome. On the right-hand side of this figure is the expected laboratory-based scenario where metabolomics data are a standalone set of information which may be frequently linked to other ‘omics data. These measurements are detailed and thus slow and usually reserved for the initial diagnostics often when disease is already apparent. This provides useful but limited retrospective information about a population. By contrast the left-hand side illustrates the role of self-testing at home which can occur much more frequently, and for some wearable devices constantly and in real-time. For example, using dipstick tests for diabetes may be a quicker assessment of glucose levels but as is already known by individuals with Type I diabetes lacks real-time prolonged monitoring of patient health. As mentioned above implantable devices are now available for real-time glucose sensing and when combined with a ‘health band’ which reports information on a patients sleep patterns, heart rate, and physical exercise schedules may lead to better management of the disease.

31The future of metabolomics does not stop at personalised medicine itself. For the application of metabolomics in preventive medicine as well as screening, the world is your oyster. Indeed, metabolomics could play not only a crucial role in monitoring life on the Earth but also beyond. NASA's recent famous twin study which was concluded last year will hopefully show a glimpse of how powerful and useful understanding the human metabolome can be,.

32At present metabolomics is very much research laboratory-based and needs to move out of academic laboratories and into the clinic. As a step towards this the UK has established two Phenome centres, one in London and the other in Birmingham; time will tell whether these are successful but a real opportunity is presented for the large-scale use of metabolomics for preventive health care, disease diagnosis, disease monitoring as well as finding novel therapeutics on a personalised level, which will account for differences within each individual.

33A recently published white paper demonstrates the strengths of metabolomics in shaping precision medicine, and we would urge all readers to dip into the text along with the accompanying Topical Issue published in Metabolomics on “Recent advances in Pharmacometabolomics: enabling tools for precision medicine”.

34As the ancient proverb says:“Vita brevis, ars longa, occasio praeceps, experimentum periculosum, iudicium difficile”

35“Vita brevis, ars longa, occasio praeceps, experimentum periculosum, iudicium difficile”

36which translates to:“Life is short, and art long, opportunity fleeting, experimentations perilous, and judgement difficult.”

37“Life is short, and art long, opportunity fleeting, experimentations perilous, and judgement difficult.”

38Thus, there is an urgent and somewhat imminent need for precision medicine! This will require appropriate infrastructure for metabolomics for (and indeed on) the masses and will require alterations in healthcare practices across the globe. Once delivered this may improve medicine, put the patient at the centre of the analysis, and allow for healthier lifestyles and efficient medication for each and every one of us.